Spatially explicit regional flood frequency analysis using L-moment, GIS and geostatistical methods.

Abstract

Lacking efficient means of digital data storage and processing, previous regional flood frequency analyses seldom considered the spatial dimension of the problem explicitly. As a result, insufficient use was made of spatial autocorrelation; of associations with other variables or of topological inter-relationships such as proximity (near a water body), containment (within a $2\rm\sp{nd}$ orderwatershed) and orientation (bearing and azimuth). Spatially-implicit methods also tended to marginalise the importance of flood-generating mechanisms--which vary continuously in space--and which combine at each location to yield the total flood risk. In this study, the most current nonparametric and L-moment (parametric) methods for flood frequency analysis (FFA) are supplemented by geostatistical and GIS techniques. Three spatial approaches are adapted for FFA: scientific visualization of random fields; characterization of spatial associations; and hierarchical spatial models of flood parameters. Three spatial models are investigated for the L-moments of flood observations: the L-skew is taken as an average within regions (polygons); the L-coefficient of variation is modelled using kriging (continuous space); and the L-mean is estimated locally (point) or from maps if local records are unavailable. L-moments are used to estimate the parameters probability distributions used to describe floods. Average daily maximum (AM) flood flows for Central and Eastern Canada are analysed. (Abstract shortened by UMI.)